Ling 3.0 Flash becomes top-performing open model in its size class

Ant Group's inclusionAI has released Ling 3.0 Flash, an open-source model under an MIT license. Scoring 38 points on the Artificial Analysis Intelligence Index, it is the top-performing open model under 124 billion parameters, matching Qwen3.6 27B with fewer active parameters. The model significantly reduces its hallucination rate on the AA Omniscience test from 97% to 44% and offers superior per-token pricing compared to competitors. It is available on Hugging Face, DeepInfra, and the inclusionAI API.

Ling 3.0 Flash performance benchmarks

• ▪Ling 3.0 Flash scored 38 points on the Artificial Analysis Intelligence Index, making it the highest-scoring open model under 124 billion total parameters as of August 13, 2026.
• ▪Ling 3.0 Flash demonstrated performance gains in agentic tasks over its predecessor, including on the t3-Bench Banking benchmark.
• ▪Ling 3.0 Flash trails the leading open model, DeepSeek V4 Flash, which scored 52 points on the Artificial Analysis Intelligence Index as of August 13, 2026.
• ▪Ling 3.0 Flash's score of 38 points on the Artificial Analysis Intelligence Index puts it on par with Qwen3.6 27B while using fewer active parameters.

Hallucination rate reduction

• ▪Ling 3.0 Flash refuses to answer questions for which it lacks reliable answers more frequently than its predecessor model.
• ▪On the Artificial Analysis Omniscience test, the hallucination rate of Ling 3.0 Flash dropped to 44 percent, down from 97 percent in its predecessor model.

Per-token pricing comparison

• ▪On per-token pricing, Ling 3.0 Flash is cheaper than any comparably capable model as of August 13, 2026.
• ▪Ling 3.0 Flash remains cheaper on a per-task basis than Qwen3.6 27B despite burning through more tokens on complex tasks.

Model release under MIT license

• ▪Ant Group's inclusionAI released Ling 3.0 Flash under an open-source MIT license on August 13, 2026.
• ▪The model weights for Ling 3.0 Flash are available directly to the public on Hugging Face, and the model is accessible via the inclusionAI API and DeepInfra.